.png){kind=icon}
.png){kind=icon}
Return
GNS2CCL: A graph network semi-supervised concept-cognitive learning model for node classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.patcog.2025.112958.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Semi-supervised node classification is a crucial research issue in graph neural networks (GNNs). While deep learning models often perform well in classification, they are usually seen as black box models, making it difficult to understand their decision-making process. This is especially unfavorable in certain application scenarios such as medical diagnosis, as people expect to have a clear understanding of why a specific outcome is predicted. However, the concept-cognitive learning has shown its advantages, such as powerful interpretability, conformity to human brain cognition, and good generalization performance. Inspired by the advantages of concept-cognitive learning, we establish a graph concept-cognitive learning model to achieve the task of graph network semi-supervised node classification. Specifically, we first propose the local graph network granular concept, and then discuss the issue of the dynamic update of this concept. Furthermore, a graph network semi-supervised concept-cognitive learning model is formulated for node classification. This model comprises the construction of concept space, pseudo-labeled process and node classification process, maintaining the advantage of good interpretability. Finally, 20 real graph network datasets are selected for controlled experiments, in which the proposed model is compared with the classical semi-supervised concept-cognitive learning model and the classical semi-supervised node classification models in GNNs, and the dynamic learning ability of the proposed model is also evaluated. The experimental results show that the proposed model can not only improve the classification accuracy, but also shares satisfactory robustness and dynamic learning ability.
Keywords:
Semi-supervised node classification
Graph network concept-Cognitive learning
Dynamic learning
Local graph granular concept